efficient performance based design using parallel...
TRANSCRIPT
UNIVERSITY OF BRITISH COLUMBIADEPARTMENT OF CIVIL ENGINEERING
C.E. VENTURA1, A. BEBAMZADEH2
EFFICIENT PERFORMANCE BASED DESIGN USING PARALLEL AND CLOUD COMPUTING
Outline
1. Why HPC in Structural Engineering
2. What is High Performance Computing (HPC)
– Cloud Computing
– Parallel Computing
3. Application of HPC in Performance-based Design
4. What is next
Why HPC in SE?
Capacity Design: Well-defined inelastic
behavior
Performance-based Design:
• Serviceability to frequent earthquakes
• Low probability of collapse during an
extremely rare event
Well-developed computer model to get a best-estimate
response at the various levels of earthquake demands
Large-scale simulations requires time, computer resources,
data processing
Speed up by high-performance computing resources, such
as multicore, GPUs, clusters, and grid and cloud services.
High Performance Computing (HPC): aggregating
computing power to solve large problems
What is HPC?
• Clusters • Multicore
computers
• Grids • Clouds
Parallel Programming is essential in HPC computing to
take the advantage of computer resources.
Lack of parallel programming experience by the software
community
What is Cloud Computing?
Cloud computing concept dates back to the
1950s . Large-scale Mainframe where used
by academia and corporations allowing
multiple users share both the physical access
to the computer from multiple terminals as
well as to share the CPU time.
Mainframe IBM 7090's at NASA's Project Mercury, 1962
Cloud Computing
High-capacity networks, low-cost
computers and storage devices led to a growth in cloud computing.
Cloud is a metaphor for the Internet
providing access to resources are though
web-based tools and applications.
Low Cost: reduce the cost by pay-as-you-go based on
demands, no need to spend big money on hardware, software or
licensing fees, no hardware maintenance
Elasticity: add and remove compute resources to meet the size
and time requirements commensurate with the size and type of
building model being analyzed.
Virtualization: applications can be easily migrated from one
physical server to another
Run Jobs Anytime, Anywhere: enable users to access
systems using a web browser regardless of their location or
what device they are using. Access in minutes instead of
spending time in queues of clusters and grids.
Why Cloud Computing?
Example of Cloud Services
NEEShub: Cloud Platform as a
Service (PaaS) for research and
education in earthquake engineering.
OpenSees are remotely run on the NEEShub cloud based
machines (McKenna et al. 2013). Parallel applications such as
OpenSeesSP and OpenSeesMP in NEEShub provides high
performance computing tools for large models or repetitive runs.
Amazon EC2 : Cloud Infrastructure as as
Service (IaaS) that provides resizable
compute capacity in the cloud.
Amazon EC2 provides flexibility to meet your computing needs
by choosing from high memory CPU instances; large CPU and
GPU clusters; and high storage instances.
Purpose: implementation of a
methodology to take full advantage of
high-performance parallel computing
using the cloud architecture for various
structural and geotechnical programs
Implementation of SE/GE Applications in Cloud Computing - UBC
Using commercial cloud
services such as Amazon EC2
to rent pay-as-you-go virtual
computers giving engineer in
design office very high CPU
capabilities
loading the model information
into the cloud controller, select the cluster types and nodes, and transfer the results.
SE/GE Application
SE/GE Applications in Cloud Computing
Applications related to seismic response of structures, such as
• Sensitivity analyses in the selection of ground motions
• Incremental dynamic analysis for low- and high-rise buildings
• Risk-based calculations of the response of various types of
buildings
• Estimation of damage and losses in buildings to various types
of earthquake mechanisms
• Ground motion directionality effects on the response to tall
buildings
• Design optimization and reliability analysis
• Soil-structure interaction in bridges
• Software
• OpenSees
• CANNY
• SAP2000 (in progress)
Example of HPC Computing in PBD
Use FWT 53-storey office building in
downtown Los Angeles to demonstrate
methodology for non-linear response
history analysis (RHA) using cloud
parallel computing.
Steel frame office tower with five levels
of underground parking. The FWT was
designed in 1988, constructed in 1988-
1980, and instrumented by CSMIP in
1990.
Structural system consists of three
main components: a braced-core,
twelve columns, and eight deep
outrigger beams at each floor
Example of HPC Computing in PBD
3-D model developed by Ventura and Ding (2000) using
CANNY and subsequently by Kalkan and Chopra
(2012) using OpenSees.
Building was modeled as a combination of non-linear
braced frames and moment frames consisting of 58
separate columns types and 23 different beam types.
The periods and response were verified with recorded
motions during the Northridge and Chino-Hills
earthquakes.
Ground motions were selected
using the Modal Pushover-
based Scaling (MPS) method
(Kalkan and Chopra 2012).
Example of HPC Computing in PBD
3D model has been modified to perform parallel
non-linear RHA using the OpenSeesSP and
OenSeesMP platforms.
High memory cluster
instances of the Amazon
EC2 Cloud Center.
• 244 GiB of memory
• 2 x Intel Xeon E5-2670
(hyper threading, eight-
core. Intel Turbo, NUMA)
≅ 32 process units
• 2.97$ / hour (Windows),
2.40$/hour (Linux)
Example of HPC Computing in PBD
Chi-Chi Taiwan
3 times reduction in run time using 4 processors. No significant
reduction using more than 6 processors due to the complexity of
model of the building and the order of parallel computation in
the model.
Amdahl's law
P: number of processora : fraction of non-parallelizable parts
Example of HPC Computing in PBD
3.5 times reduction in run time for
set of 6 ground using 6 processors,
in which each ground motion was
executed by one processor and
Up to 9 times by using 18
processes and assigning 3
processors for each ground motion
run for about 5$
Incremental Dynamic Analysis
for a set of 6 ground motions at 8
levels of intensity
Set of scaled ground motions
Incremental Dynamic analysis
The runtime can be reduced from
78 hours using only one processor
to about 1.5 hour using 8 high
memory EC2 clusters instances with
total number of 128 processors for
about 50$
Example of HPC Computing in IDA
0 1 2 30
10
20
30
40
5053
Floor Displacement/Building Height (%)
Flo
or
25% MPS
50% MPS
75% MPS
100% MPS
125% MPS
150% MPS
175% MPS
200% MPS
0 5 100
10
20
30
40
5053
Drift (%)
Flo
or
25% MPS
50% MPS
75% MPS
100% MPS
125% MPS
150% MPS
175% MPS
200% MPS
0 2 4 6 8 100
25
50
75
100
125
150
175
200
Drift (%)
% M
PS
GM 1
GM 2
GM 3
GM 4
GM 5
GM 6
• Median of displacement and drift of 6 ground motions at each floor of
52-storey building and at different percentage of MPS factors
• Drift at 15th floor for set of
6 ground motions and
different percentage of
MPS factors
HPC Computing in Ground Motion Directionality
How different could the calculated response of a high-rise building
structure be when the directionality of the ground motion is
considered?
HPC Computing in Ground motion directionality
7 times reduction in run time
for set of 6 ground using 7
processors, in which each
ground motion angle was
executed by one processor and
Up to 17 times by using 21
processes and assigning 3
processors for each ground
motion run.
Unscaled Northridge ground
motion at 7 different direction
angles were considered: 0,
15, 30, 45, 60, 75, and 90
degrees.
0 1 2 30
10
20
30
40
5053
Drift (%)
Flo
or
0o angle
Largest response
Seismic Performance Analyzer I
Analyzer I is a web-based application that gives the engineer access to a large database of non-linear dynamic analysis results for a comprehensive parametric range of structural element types in Low-rise buildings.
Engineers are relieved from performing sophisticated nonlinear analysis for individual buildings in order to benefit from the advantages of probabilistic performance-based design
approach.
Seismic Performance Analyzer 2
Analyzer Version 2.0 is a high-performance seismic analysis tool for the seismic risk assessment and retrofit design of low-rise and mid-rise buildings.
1) Selection of ground motions / SRG/FEMA-P6952) Shearwalls /moment frames3) Incremental dynamic analysis 4) Fast and computationally efficient manner using
parallel and cloud computing
Seismic Performance Analyzer 2
1) Easy defining model properties2) Performing IDA for different region and soil types
Seismic Performance Analyzer 2
1) Capabilities of parallel and cloud computing2) 750 nonlinear time history analyses in less than 7 min
Seismic Performance Analyzer 2
1) Calculating probabilities of drift and plastic rotations2) Base shear and overturning moment
The examples presented clearly illustrated the advantages of using parallel cloud computing for the dynamic analysis of a tall building.
The significant reduction in the time required to process a suite of ground motions, and the flexibility of cloud computing in assigning tasks to different processor, allow engineers to perform various types of analyses in parallel.
Proposed methodology provides a tool that enables structural engineers to run high performance computer (HPC) applications such as performance-based design, IDA, directionality,
optimization, and SSI very efficiently
The main advantages are: low cost: elasticity: run jobs
anytime, anywhere without spending time in queues.
In Summary
• British Columbia Schools Seismic Retrofit Program.
• British Columbia Ministry of Education
• Association of Professional Engineers and Geoscientists of British
Columbia (APEGBC)
• University of British Columbia; the APEGBC Structural Peer Review
Committee (BC engineers); and the APEGBC External Peer Review
committee (California engineers).
• The authors express their thanks to Drs. Farzad Naeim, Michael
Mehrain and Robert Hanson.
• Erol Kalkan from the USGS.
Acknowledgment